Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74848.77130094705,
-      "required_pages_per_second_upper_estimate": 0.006487701793889572
+      "required_token_positions_per_second": 75821.11651308216,
+      "required_pages_per_second_upper_estimate": 0.006571982212491034
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9356.09641261838,
-      "required_pages_per_second_upper_estimate": 0.0037322867451006787
+      "required_token_positions_per_second": 9477.63956413527,
+      "required_pages_per_second_upper_estimate": 0.003780772125472822
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18712.19282523676,
-      "required_pages_per_second_upper_estimate": 0.00432873648677259
+      "required_token_positions_per_second": 18955.27912827054,
+      "required_pages_per_second_upper_estimate": 0.004384970224806639
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28068.289237855144,
-      "required_pages_per_second_upper_estimate": 0.0017092115516528263
+      "required_token_positions_per_second": 28432.91869240581,
+      "required_pages_per_second_upper_estimate": 0.001731415572372063
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28068.289237855144,
-      "required_pages_per_second_upper_estimate": 0.0021607746035275923
+      "required_token_positions_per_second": 28432.91869240581,
+      "required_pages_per_second_upper_estimate": 0.0021888447882978347
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28068.289237855144,
-      "required_pages_per_second_upper_estimate": 0.34851421381296976
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28432.91869240581,
+      "required_pages_per_second_upper_estimate": 0.35304169130220653
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.14907782916378665,
-    "producer_p95_seconds": 0.3851987430025474,
-    "gpu_update_seconds": 0.9850261897227314,
-    "producer_batches_per_second": 6.707905565899639,
-    "gpu_batches_per_second": 1.0152014336608486,
-    "producer_headroom": 6.607462660598023,
+    "producer_batch_seconds": 0.11330460203873827,
+    "producer_p95_seconds": 0.376073257997632,
+    "gpu_update_seconds": 0.97239401621419,
+    "producer_batches_per_second": 8.825766844475611,
+    "gpu_batches_per_second": 1.0283897096500945,
+    "producer_headroom": 8.582122868069678,
     "sustainable": true,
-    "buffer_seconds": 60.086597573086614,
+    "buffer_seconds": 60.28842900527978,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 158,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 573411.5990591539,
-    "upload_bytes_per_second": 5697933.396816317,
+    "generation_bytes_per_second": 551422.2772046723,
+    "upload_bytes_per_second": 5780483.223379054,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.52039040961725,
+    "estimated_full_shards_per_hour": 118.32238423447731,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.143835795999621,
-      "bytes": 1177702,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.4552070569989155,
       "bytes": 11480592,
@@ -562,6 +556,12 @@
       "bytes": 2305951,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.9167553060033242,
+      "bytes": 5839408,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -337,3 +337,4 @@
 {"step": 228700, "loss": 0.593542791903019, "lr": 0.0003, "grad_norm": 0.15870941771635896, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33475266402, "tokens_per_second": 149403.2379047505, "data_wait_s": 1.26349987112917e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 228800, "loss": 0.5507429484277964, "lr": 0.0003, "grad_norm": 0.14181171220754413, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33489760417, "tokens_per_second": 146354.72225646026, "data_wait_s": 9.732000762596726e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 228900, "loss": 0.6200902182608843, "lr": 0.0003, "grad_norm": 0.1488659235463212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33504225219, "tokens_per_second": 147854.12281896625, "data_wait_s": 9.461997251491994e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 229000, "loss": 0.5841774679720402, "lr": 0.0003, "grad_norm": 0.15640982062436973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33518730851, "tokens_per_second": 146847.44436192018, "data_wait_s": 1.1139003618154675e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:db66dc8c693c17381ba360ebf7265d1d18582af36d493ed5a6b17fdb56ba3f87
-size 722865
+oid sha256:16f8afbb8e1204173f9ad86eddf8d42532fd48c2f83871b87e78c4b77b7adf52
+size 724983